Bug 14538 - [core TTM] Premature fence timeouts.
Summary: [core TTM] Premature fence timeouts.
Status: NEW
Alias: None
Product: DRI
Classification: Unclassified
Component: DRM/other (show other bugs)
Version: DRI git
Hardware: Other All
: medium normal
Assignee: Default DRI bug account
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-02-18 02:32 UTC by Thomas Hellström
Modified: 2008-02-18 02:32 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Thomas Hellström 2008-02-18 02:32:46 UTC
Symtoms:
gltestperf times out with an -EBUSY (-16).

It's possible to feed enough rendering commands in a command FIFO for the GPU to be busy more than 3 seconds, which is the default fence wait.

Proposed fix:
1) Add an issued_jiffies timestamp member to the fence. The fence maximum wait time will be relative to this timestamp, instead of relative to the wait start. This will remove all needs for extra user-space fence time-out checks.

2) If a timeout is hit, we call a device-specific callback "fence_timeout" that checks in one way or another whether the GPU is still alive. In that case, reset issued_jiffies to the current time. If not, Really time out.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.